Optical Lattice Clocks

A caesium clock counts the ticks of a microwave hyperfine transition at 9.2 GHz. An optical lattice clock counts a transition a hundred thousand times faster — an optical line at hundreds of terahertz — and that single change is the largest lever on clock performance ever pulled, because for a fixed linewidth the quality factor and the achievable stability both scale with the transition frequency. But a fast reference is only as good as how long and how gently you can watch it: probe too briefly and the Fourier-limited line is broad; watch a free atom too long and it Doppler-shifts and drifts out of the beam. Trapping fixes the time budget, yet every trap that holds a neutral atom holds it through the AC Stark shift — a second-order perturbation of the atomic levels, borrowed straight from quantum mechanics — which by construction moves the very clock frequency you are trying to measure. The optical lattice clock threads this needle with two ideas layered on the dipole trap, the recoil limit, and the sub-Doppler cooling this chapter has already built. First, the magic wavelength: tune the lattice laser to a colour where the two clock states have equal dynamic polarizability, so both levels sink by the same energy and the gap between them — the thing the clock reads — does not move, even in a deep trap. Second, the Lamb–Dicke regime: confine each atom to a site smaller than the clock wavelength, so the photon recoil is absorbed by the whole lattice, Mössbauer-style, leaving a recoil- and first-order-Doppler-free carrier. Add thousands of atoms interrogated at once, and projection noise averages as one over the square root of the atom number. The result reaches the 10⁻¹⁸ decade, where the height of the apparatus above the geoid becomes a limiting effect.

Key takeaways

An optical lattice clock reads a narrow optical transition — the ¹S₀ → ³P₀ line of strontium or ytterbium at hundreds of THz — in thousands of atoms held in a standing-wave dipole trap, exploiting the enormous line quality factor Q = ν₀/Δν.
The trap light unavoidably Stark-shifts both clock levels, but at a magic wavelength the ground and excited dynamic polarizabilities are equal, so the differential light shift — the clock error — cancels to first order in intensity.
Confinement in the Lamb–Dicke regime (site size x₀ ≪ clock wavelength) hands the photon recoil to the lattice, giving a recoil- and first-order-Doppler-free carrier flanked by resolved motional sidebands.
Interrogating N ≈ 10³–10⁴ atoms at once beats down quantum projection noise as 1/√N — the lattice clock's structural edge over a single-ion clock at equal Q.
State-of-the-art Sr and Yb clocks reach instability and inaccuracy at or below 10⁻¹⁸, making them the leading candidates to redefine the SI second and enabling centimetre-scale relativistic geodesy.
Key idea: trap many atoms in a standing wave tuned to the magic wavelength where the two clock states share the same AC polarizability, so the lattice confines them in the Lamb–Dicke regime — killing Doppler and recoil shifts — without shifting the clock transition, yielding a recoil-free, light-shift-cancelled optical reference at the 10⁻¹⁸ level.

The core insight

Trap many atoms in a standing wave tuned to the magic wavelength, where the two clock states have equal AC polarizability, so the trap confines them deep in the Lamb–Dicke regime — eliminating Doppler and recoil shifts — without moving the clock transition, producing a recoil-free, light-shift-cancelled optical reference accurate to 10⁻¹⁸.

Where this lives today

The flagship instruments are the strontium and ytterbium lattice clocks at JILA, NIST, PTB, RIKEN, SYRTE and NMIJ, several now agreeing at or below the low-10⁻¹⁸ level. A modern Sr clock is room-sized: a blue-then-red magneto-optical trap loads ⁸⁷Sr, atoms move into a lattice at 813 nm — increasingly a 2D, 3D, or tweezer-array lattice — inside a well-characterised thermal enclosure, and an ultrastable laser pre-stabilised to a cryogenic silicon cavity interrogates the 698 nm line for up to a second per cycle. A femtosecond comb divides that optical frequency to a countable microwave rate and phase-links it to other clocks over stabilised fibre. The comparisons already have teeth: clocks on different floors have resolved the redshift across tens of centimetres, transportable Sr clocks have measured height differences by time alone, and Sr–Yb–Al⁺ comparisons bound any drift of the fine-structure constant and hunt for ultralight dark matter. The next redefinition of the second will almost certainly retire caesium for one of these optical transitions.

The physical picture

Picture the trap laser as a hand pressing down on both clock levels. Ordinarily it presses them by different amounts, so the gap between them — the clock frequency — slides as you change the trap power. That is a catastrophe: the trap you need for a long measurement is the very thing corrupting the answer. The magic wavelength is the colour at which the hand presses both levels down by exactly the same amount. Each level is shifted enormously — tens of kilohertz — yet the difference, all the clock cares about, sits perfectly still even as the atom rides deep in a strong trap where the intensity varies from site to site and across each site.

Figure

The magic wavelength for strontium near 813.4 nm: the ground-state polarizability α_g and excited-state polarizability α_e cross, so the differential light shift Δα = α_e − α_g passes through zero and the clock frequency stops depending on trap power

Now freeze the motion. An atom in a lattice well is a tiny quantum oscillator whose ground-state wavepacket has an r.m.s. size x₀ far smaller than the clock wavelength it is about to absorb. When it absorbs, the momentum kick ħk that would normally recoil-shift and Doppler-broaden the line cannot change the atom's motional state — there is not enough room to move — so the recoil is taken up by the whole lattice, exactly as a gamma photon's recoil is taken by a crystal in the Mössbauer effect. The spectrum collapses into a sharp central carrier at the true atomic frequency, with the vibrational quanta appearing as symmetric sidebands offset by the trap frequency. The second figure shows both halves: the tightly held wavepacket, and the recoil-free carrier with its resolved sidebands.

Figure

Left: each atom's wavepacket of size x₀ sits far inside one lattice well, x₀ ≪ λ_clock, giving η ≈ 0.26. Right: the resulting spectrum is a recoil-free carrier at ν₀ flanked by resolved motional sidebands at ±ν_trap ≈ 69 kHz for a depth of s = 100 recoils

Stability is then a counting game: with N atoms per cycle the instability falls as one over Q times the square root of one over N, so optical Q near 10¹⁵ and N near 10⁴ together buy a stability a single trapped ion cannot match by statistics alone.

The core equations

An atom in a light field acquires an energy shift because the field induces a dipole that then interacts with the field that made it. That shift is the trapping potential — to second order in the field, proportional to the atom's dynamic polarizability and the local intensity. The minus sign for a red-detuned trap makes intensity maxima attractive.

U(\mathbf{r},\omega)=-\tfrac12\,\alpha(\omega)\,\langle E^2(\mathbf{r})\rangle=-\frac{1}{2\varepsilon_0 c}\,\alpha(\omega)\,I(\mathbf{r}) \qquad (1)

where: $U$ — dipole (light-shift) potential energy; $\alpha(\omega)$ — dynamic (AC) scalar polarizability at trap frequency $\omega$ ; $\langle E^2\rangle$ — time-averaged square of the field; $I$ — local intensity; $\varepsilon_0,\,c$ — vacuum permittivity and speed of light. Reading it: the trap depth is set entirely by the polarizability, so two states with different $\alpha$ sit in different-depth traps — the enemy. Subtracting the two state shifts gives the clock-frequency error, proportional to the differential polarizability. The whole design turns on the bracket below, because it can be driven to zero without turning off the intensity.

h\,\Delta\nu_{\mathrm{clock}}=U_e-U_g=-\frac{1}{2\varepsilon_0 c}\big[\alpha_e(\omega)-\alpha_g(\omega)\big]\,I\;\xrightarrow{\ \alpha_e(\omega_m)=\alpha_g(\omega_m)\ }\;0 \qquad (2)

where: $\Delta\nu_{\mathrm{clock}}$ — shift of the $g\!\to\!e$ frequency; $\alpha_{e},\alpha_{g}$ — polarizabilities of the excited and ground clock states; $h$ — Planck constant; $\omega_m$ — magic frequency, $\lambda_m=2\pi c/\omega_m$ . Reading it: the clock reads only the difference of the two light shifts, so a crossing of the two polarizabilities nulls the shift for every atom regardless of its place in the intensity profile. Whether the recoil trick works is decided by one dimensionless number — the ratio of wavepacket size to clock wavelength, equivalently the square root of the recoil energy over the vibrational quantum. It must be much less than one.

\eta=k\,x_{0}=k\sqrt{\frac{\hbar}{2m\,\omega_{\mathrm{trap}}}}=\sqrt{\frac{E_r}{\hbar\,\omega_{\mathrm{trap}}}} \qquad (3)

where: $\eta$ — Lamb–Dicke parameter; $k=2\pi/\lambda_{\mathrm{clock}}$ — clock (probe) wavevector; $x_0$ — ground-state r.m.s. extent of the well; $\omega_{\mathrm{trap}}$ — vibrational trap frequency; $E_r=\hbar^2k^2/2m$ — clock-photon recoil energy; $m$ — atomic mass. Reading it: small $\eta$ means one vibrational quantum costs far more than one recoil, so the recoil cannot excite motion and is shunted to the lattice. To evaluate $\eta$ you need the trap frequency, from expanding the standing-wave potential about a minimum. Measuring the depth in units of the lattice recoil energy $s$ makes the trap frequency a tidy multiple of the recoil frequency.

U_0=s\,E_r^{(L)},\qquad \omega_{\mathrm{trap}}=k_L\sqrt{\frac{2U_0}{m}}=2\sqrt{s}\,\frac{E_r^{(L)}}{\hbar} \qquad (4)

where: $U_0$ — trap depth; $s$ — depth in lattice recoils; $E_r^{(L)}=\hbar^2k_L^2/2m$ — lattice-photon recoil energy; $k_L=2\pi/\lambda_L$ — lattice wavevector. Reading it: deeper lattices (larger $s$ ) tighten the well, raise $\omega_{\mathrm{trap}}$ , and push $\eta$ down only as $s^{-1/4}$ — while raising the lattice light shift, so $s$ is a compromise. Given a recoil-free carrier of quality factor $Q$ , the fundamental noise floor is quantum projection noise: each atom's spin readout is a coin flip, and averaging over $N$ atoms and many cycles beats it down. This is the formula every clock is judged against.

\sigma_y(\tau)=\frac{1}{\pi Q}\sqrt{\frac{T_c}{N\,\tau}},\qquad Q=\frac{\nu_0}{\Delta\nu} \qquad (5)

where: $\sigma_y(\tau)$ — Allan deviation at averaging time $\tau$ ; $Q$ — line quality factor; $\nu_0$ — clock frequency (≈ 4.3 × 10¹⁴ Hz for Sr); $\Delta\nu$ — Fourier-limited linewidth; $T_c$ — cycle time; $N$ — atoms per cycle. Reading it: stability improves with a bigger $Q$ (optical over microwave) and with more atoms (the lattice over the single ion). With light shift and motion tamed, the leading systematic is the ambient thermal field: room-temperature blackbody radiation is a broadband, isotropic field producing a DC-like differential Stark shift that grows as the fourth power of temperature.

\Delta\nu_{\mathrm{BBR}}=-\frac{1}{2h}\,\Delta\alpha_0\,\langle E^2\rangle_T,\qquad \langle E^2\rangle_T=\frac{1}{\varepsilon_0}\,\frac{\pi^2 (k_BT)^4}{15\,\hbar^3 c^3} \qquad (6)

where: $\Delta\nu_{\mathrm{BBR}}$ — thermal-field clock shift; $\Delta\alpha_0$ — differential static polarizability ( $\omega\!\to\!0$ ); $\langle E^2\rangle_T$ — mean-square thermal field; $k_B,\,T$ — Boltzmann constant and ambient temperature. Reading it: because the shift grows as $T^4$ , its uncertainty grows as $4\,\delta T/T$ , so the temperature must be known to millikelvins to correct it at $10^{-18}$ . Finally, at $10^{-18}$ general relativity is a signal, not a correction. Two identical clocks at different heights tick at different rates through the gravitational redshift, $\Delta\nu/\nu=g\,\Delta h/c^2\approx1.09\times10^{-16}$ per metre (with $g\approx9.81\ \mathrm{m\,s^{-2}}$ and $c$ the speed of light) — one centimetre of height is a part in $10^{18}$ , so a $10^{-18}$ clock resolves a centimetre of elevation through time alone, turning it into a probe of the geopotential.

Derivation AWhy the magic wavelength cancels the light shift

The goal is to show that the clock-frequency error is just the difference of two intensity-proportional AC Stark shifts, and that because the two clock states couple to different levels, their dynamic polarizabilities cross at discrete wavelengths — where the error vanishes identically in the intensity, and so is nulled everywhere in the trap at once, even in the deep, intensity-varying centre of a site.

Step 1. For each state $i\in\{g,e\}$ , second-order perturbation theory gives $U_i(\omega)=-\tfrac12\,\alpha_i(\omega)\,\langle E^2\rangle$ with $\alpha_i(\omega)=\sum_k \dfrac{2\omega_{ki}\,|\langle k|d|i\rangle|^2}{\hbar(\omega_{ki}^2-\omega^2)}$ , a sum over dipole couplings to all other levels $k$ . Why: the resonance structure of this sum is what makes $\alpha_g$ and $\alpha_e$ genuinely different functions of $\omega$ .

Step 2. Subtract to get the clock error, $h\,\Delta\nu_{\mathrm{clock}}=U_e-U_g=-\tfrac12[\alpha_e(\omega)-\alpha_g(\omega)]\langle E^2\rangle$ . Why: the clock never sees either level's absolute shift, only the gap, so the entire systematic is carried by $\Delta\alpha(\omega)=\alpha_e-\alpha_g$ .

Step 3. Because the two curves cross at discrete $\lambda_m$ , there $\Delta\alpha(\omega_m)=0$ and the bracket vanishes for every value of $\langle E^2\rangle$ . Why: this is the crucial move — the null is intensity-independent, so it holds at the bottom of a deep well and in a shallow one alike, with no need to make the trap uniform.

Step 4. What survives: the hyperpolarizability shift (∝ I²), the vector shift (∝ ellipticity and magnetic sublevel), and the tensor shift (∝ probe polarization). Why: the crossing kills only the leading scalar term, so these residuals — plus sub-picometre knowledge of $\lambda_m$ — set the true lattice budget. *Verified: reduces to zero at the crossing; consistent with Katori's magic-wavelength proposal, Phys. Rev. Lett. 91, 173005 (2003).*

Go deeper · Derivation B · 1 minLamb–Dicke suppression of recoil and first-order Doppler

Derivation BLamb–Dicke suppression of recoil and first-order Doppler

The goal is to prove that a tightly bound atom absorbs a clock photon without recoiling: the recoil phase factor, written in the trap's ladder operators, is a displacement operator whose diagonal element — the carrier — tends to unity as $\eta\to0$ , so the motional state is unchanged and the kick goes to the lattice, while the off-diagonal elements become the motional sidebands.

Step 1. Write the recoil phase as $e^{ikz}$ and substitute $z=x_0(a+a^\dagger)$ , giving $e^{ikz}=e^{i\eta(a+a^\dagger)}$ , a displacement of size $\eta=kx_0$ . Why: in a trap the coordinate $z$ is an oscillator variable, so the recoil literally is a coherent displacement of the motional state by $\eta$ .

Step 2. The exact carrier amplitude ( $n'=n$ ) is $\langle n|e^{i\eta(a+a^\dagger)}|n\rangle=e^{-\eta^2/2}L_n(\eta^2)\xrightarrow{\eta\ll1}1-\eta^2(n+\tfrac12)+\mathcal{O}(\eta^4)$ . Why: to leading order the carrier keeps full strength and does not move the atom — the hallmark of the Lamb–Dicke regime.

Step 3. Because the carrier preserves $n$ , the atom's momentum is unchanged: $\hbar k$ is delivered to the lattice and apparatus. Why: this removes the photon-recoil shift ( $E_r/h\sim$ kHz, otherwise a fractional systematic near $10^{-11}$ ) and the first-order Doppler shift, just as a crystal absorbs a gamma recoil.

Step 4. The terms $n'=n\pm1$ appear at $\pm\omega_{\mathrm{trap}}$ with amplitude of order $\eta$ ; when $\omega_{\mathrm{trap}}$ exceeds every linewidth (resolved-sideband regime) they separate cleanly from the carrier. Why: the ratio of red to blue sideband heights measures the mean occupation $\bar n$ , giving an in-situ thermometer. *Verified: recovers the Dicke/Mössbauer narrowing result; the carrier is recoil- and first-order-Doppler-free for $\eta\ll1$ .*

Go deeper · Derivation C · 1 minTrap frequency and the Lamb–Dicke parameter from a standing wave

Derivation CTrap frequency and the Lamb–Dicke parameter from a standing wave

The goal is to connect the two knobs an experimenter turns — lattice wavelength and dimensionless depth $s$ — to the number that decides whether the clock works, $\eta$ . We expand a $\sin^2$ site into a harmonic well, read off the vibrational frequency, and fold it into $\eta$ , which shrinks only as $s^{-1/4}$ , so modest depths already reach the Lamb–Dicke regime.

Step 1. Model one site as the minimum of $U(z)=U_0\sin^2(k_L z)$ and expand: $U(z)\approx U_0 k_L^2 z^2\equiv\tfrac12 m\omega_{\mathrm{trap}}^2 z^2$ . Why: near the bottom every smooth well is parabolic, so a site is a harmonic oscillator with levels spaced by $\hbar\omega_{\mathrm{trap}}$ .

Step 2. Read off $\omega_{\mathrm{trap}}=k_L\sqrt{2U_0/m}$ ; insert $U_0=s\,E_r^{(L)}$ and $E_r^{(L)}=\hbar^2k_L^2/2m$ to get $\omega_{\mathrm{trap}}=2\sqrt{s}\,E_r^{(L)}/\hbar$ . Why: expressing depth in recoils makes the trap frequency a clean multiple of the recoil frequency, so $s=100$ gives $\omega_{\mathrm{trap}}/2\pi=20\,(E_r^{(L)}/h)$ .

Step 3. Insert into $\eta=k_{\mathrm{clock}}\sqrt{\hbar/2m\omega_{\mathrm{trap}}}$ ; the mass cancels and $\eta=\dfrac{1}{\sqrt2}\,\dfrac{\lambda_L}{\lambda_{\mathrm{clock}}}\,s^{-1/4}$ . Why: $\eta$ depends only on the wavelength ratio and a weak quarter-power of depth, so you rely mostly on a clock wavelength short compared with the lattice, not on cranking $s$ .

Step 4. For Sr ( $\lambda_L=813.4$ nm, $\lambda_{\mathrm{clock}}=698$ nm, $s=100$ ): $\eta=(0.707)(1.165)(0.316)=0.26$ , comfortably below one. Why: $\eta\approx0.26$ confirms the Lamb–Dicke regime for realistic depths, so the carrier is recoil-free while the first sideband stays weak, of order $\eta^2\approx7\%$ . *Verified: dimensionally consistent; $\eta=0.26$ matches $\sqrt{E_r/\hbar\omega_{\mathrm{trap}}}$ with $\nu_{\mathrm{trap}}\approx69$ kHz.*

Limiting cases

The lattice clock is engineered to live in several regimes at once; each row is a design inequality that must hold.

Case	Result	Meaning
Lamb–Dicke regime	$\eta=k\,x_0=\sqrt{E_r/\hbar\omega_{\mathrm{trap}}}\ll1$	the wavepacket $x_0$ is far smaller than the clock wavelength; the carrier is recoil- and first-order-Doppler-free
Magic-wavelength condition	$\alpha_g(\lambda_m)=\alpha_e(\lambda_m)$	equal dynamic scalar polarizabilities cancel the differential light shift to first order in intensity
Resolved-sideband regime	$\omega_{\mathrm{trap}}\gg\gamma_{\mathrm{probe}},\ \Gamma$	the trap frequency exceeds probe and transition linewidths, so motional sidebands separate from the carrier
Deep-lattice / harmonic limit	$U_0\gg E_r^{(L)}$ , i.e. $s\gg1$	each site is a good harmonic oscillator and tunnelling between sites is negligible
Cryogenic / BBR-free limit	$T\to0$ , $\Delta\nu_{\mathrm{BBR}}\propto T^4\to0$	cooling the radiation shield suppresses the thermal shift far faster than the signal

Common trapOnce the magic wavelength is set the lattice is perturbation-free

It is not. The magic wavelength cancels only the first-order, scalar differential light shift, and three things survive. The hyperpolarizability shift grows as the square of intensity, so it does not track the linear null and must be measured by varying trap depth. The vector and tensor shifts depend on the lattice polarization and the atomic magnetic sublevel, and reappear the moment the lattice is slightly elliptical or the bias field tilts. And the magic wavelength itself must be held to sub-picometre precision, because the slope of $\Delta\alpha$ through the crossing is not zero — sit a few picometres off and the linear light shift creeps back. A magic wavelength is a null to be maintained, not a switch flipped once.

Worked examples

1 · Trap frequency and Lamb–Dicke parameter for strontium (the confidence builder)

A ⁸⁷Sr lattice runs at the magic wavelength λ_L = 813.4 nm, depth s = 100. The clock line is at λ_clock = 698 nm. Find the lattice recoil frequency, the axial trap frequency, and the Lamb–Dicke parameter, and confirm the Lamb–Dicke regime.

Step 1. With $m=87\,u=1.445\times10^{-25}$ kg and $k_L=2\pi/813.4\ \mathrm{nm}=7.72\times10^{6}\ \mathrm{m^{-1}}$ , the lattice recoil energy $E_r^{(L)}=\hbar^2k_L^2/2m$ gives $E_r^{(L)}/h=\mathbf{3.47\ kHz}$ — the natural recoil unit.

Step 2. Trap frequency: $\nu_{\mathrm{trap}}=2\sqrt{s}\,(E_r^{(L)}/h)=2\times10\times3.47\ \mathrm{kHz}=\mathbf{69\ kHz}$ , the offset of the sidebands in the second figure.

Step 3. Lamb–Dicke parameter, using the clock recoil energy at 698 nm ( $E_r/h=4.70$ kHz): $\eta=\sqrt{E_r/\hbar\omega_{\mathrm{trap}}}=\sqrt{4.70/69}=\mathbf{0.26}$ . The closed form $\eta=(1/\sqrt2)(\lambda_L/\lambda_{\mathrm{clock}})s^{-1/4}=(0.707)(1.165)(0.316)=0.26$ agrees.

Step 4. Since $\eta=0.26\ll1$ , the carrier is recoil- and first-order-Doppler-free, and the first sideband carries probability $\sim\eta^2\approx0.07$ — weak but easily resolved because 69 kHz dwarfs the sub-Hz probe linewidth. The clock sits firmly in the Lamb–Dicke and resolved-sideband regimes: ν_trap ≈ 69 kHz, η ≈ 0.26.

2 · The quantum-projection-noise stability of a strontium clock

Take ν₀ = 4.29 × 10¹⁴ Hz, a 1 Hz Fourier-limited linewidth (Q = 4.3 × 10¹⁴), N = 10⁴ atoms, cycle time T_c = 1 s. Estimate the fundamental instability at τ = 1 s and after one hour.

Step 1. $Q=\nu_0/\Delta\nu=4.29\times10^{14}/1=\mathbf{4.3\times10^{14}}$ — fifteen orders of magnitude, the payoff of an optical reference.

Step 2. $\sigma_y(1\,\mathrm{s})=\dfrac{1}{\pi Q}\sqrt{\dfrac{T_c}{N\tau}}=\dfrac{1}{\pi\cdot4.3\times10^{14}}\sqrt{\dfrac{1}{10^4}}=(7.4\times10^{-16})(10^{-2})=\mathbf{7.4\times10^{-18}}$ .

Step 3. White frequency noise averages as $\tau^{-1/2}$ , so after one hour $\sigma_y=7.4\times10^{-18}/\sqrt{3600}=\mathbf{1.2\times10^{-19}}$ at the projection floor. Real clocks fall short of this — the Dick effect (aliased local-oscillator noise through the dead time) and residual laser noise cap real instabilities near $10^{-16}/\sqrt\tau$ , which is why cryogenic-cavity lasers and zero-dead-time interrogation matter.

Step 4. Both levers in $\sigma_y\propto1/(Q\sqrt N)$ favour the lattice: optical $Q$ beats microwave by $10^5$ , and $10^4$ atoms beat one ion by $\sqrt N=100$ . The QPN floor is ≈ 7 × 10⁻¹⁸ at 1 s, averaging toward 10⁻¹⁹ in an hour.

3 · How well must you know the temperature? (the BBR budget)

The room-temperature BBR shift of a Sr clock is Δν_BBR/ν₀ ≈ −5.5 × 10⁻¹⁵ at T = 300 K. To correct it to a residual of 1 × 10⁻¹⁸, how well must the effective radiation temperature be known?

Step 1. Since $\Delta\nu_{\mathrm{BBR}}\propto T^4$ , a temperature error propagates as $\delta(\Delta\nu_{\mathrm{BBR}})/\Delta\nu_{\mathrm{BBR}}=4\,\delta T/T$ .

Step 2. For a residual $10^{-18}$ against a shift of $5.5\times10^{-15}$ , the shift must be known to $10^{-18}/(5.5\times10^{-15})=1.8\times10^{-4}$ of itself.

Step 3. Then $\delta T/T=\tfrac14(1.8\times10^{-4})=4.5\times10^{-5}$ , so $\delta T=(4.5\times10^{-5})(300\ \mathrm{K})\approx\mathbf{14\ mK}$ .

Step 4. The radiative environment must be mapped to about 14 mK — and since the shield is never perfectly isothermal, that means well-characterised enclosures or a cryogenic environment plus a precisely measured $\Delta\alpha_0$ . Reaching 10⁻¹⁸ demands the effective BBR temperature to ≈ 14 mK.

Physics of the portrait

The optical lattice clock is the convergence of three lines of work — the narrow optical transition, the neutral-atom trap, and the frequency comb that counts optical cycles. One person turned the trap into a clock by seeing that a light shift could be engineered to cancel.

Origins · 1 minHidetoshi Katori and how this was discovered

Hidetoshi Katori

b. 1964 · magic-wavelength optical lattice clocks, University of Tokyo

For decades the received wisdom was that you must never trap the atoms you clock: any confining light shifts the levels, so the best optical standards used a single ion in an almost field-free RF trap, accepting the poor statistics of one particle. Hidetoshi Katori inverted the objection. If the trap light must shift the levels, then choose the one colour at which it shifts both clock levels equally — the "magic wavelength" — so the transition frequency is untouched while thousands of neutral atoms are held motionless in a standing wave. Working at the University of Tokyo, Katori identified the ¹S₀ → ³P₀ transition of strontium as ideal: doubly forbidden, hair-thin, and blessed with a convenient magic wavelength near 813 nm. His group's 2003 demonstration (Phys. Rev. Lett. 91, 173005) turned a theoretical trick into a working reference and launched the field. He went on to build cryogenic clocks that tame the blackbody shift and transportable Sr clocks carried up towers and mountains to measure the gravitational redshift directly — the first steps of clock-based relativistic geodesy. The magic-wavelength lattice is now the template for the best clocks on Earth, and the reason redefining the SI second is a question of when, not whether.

History

1967

— the SI second is redefined on the caesium-133 ground-state hyperfine transition

1975

— Dehmelt proposes a single trapped ion as an optical frequency standard of extreme Q

2000

— the self-referenced femtosecond frequency comb links optical and microwave frequencies

2003

— Katori and co-workers demonstrate the strontium magic-wavelength optical lattice

2014

— a JILA strontium lattice clock reaches the 10⁻¹⁸ decade in systematic uncertainty

Going further

Texts. Foot, Atomic Physics, ch. 7 and 9 (light shifts, precision spectroscopy) · Demtröder, Laser Spectroscopy (optical frequency standards and Lamb–Dicke spectroscopy) · Loudon, The Quantum Theory of Light (AC Stark shift, dressed atom) · Saleh & Teich, Fundamentals of Photonics (standing waves and dipole traps) · Ludlow, Boyd, Ye, Peik & Schmidt, "Optical atomic clocks", Rev. Mod. Phys. 87, 637 (2015).

Provenance. A verifier should confirm: (1) the AC Stark potential $U=-\tfrac12\alpha\langle E^2\rangle$ and magic condition $\alpha_e=\alpha_g$ (Foot ch. 7; Loudon; Katori, PRL 91, 173005 (2003)). (2) $\eta=kx_0=\sqrt{E_r/\hbar\omega_{\mathrm{trap}}}$ and $\omega_{\mathrm{trap}}=2\sqrt s\,E_r^{(L)}/\hbar$ from the $\sin^2$ expansion; for ⁸⁷Sr at 813.4 nm, $E_r^{(L)}/h\approx3.47$ kHz, $\nu_{\mathrm{trap}}\approx69$ kHz at $s=100$ , $\eta\approx0.26$ using the 698 nm clock recoil. (3) $\sigma_y=(1/\pi Q)\sqrt{T_c/N\tau}$ , giving $\sigma_y(1\,\mathrm{s})\approx7\times10^{-18}$ for $Q=4.3\times10^{14}$ , $N=10^4$ , $T_c=1$ s. (4) The redshift $g/c^2\approx1.09\times10^{-16}\ \mathrm{m^{-1}}$ (from $9.81/(2.998\times10^8)^2$ ), i.e. $1.09\times10^{-18}$ per cm. (5) The $T^4$ scaling of the BBR shift; the Sr value $\Delta\nu_{\mathrm{BBR}}/\nu_0\approx-5.5\times10^{-15}$ at 300 K and the ≈ 14 mK budget are standard. Other numbers (Sr $\nu_0\approx4.3\times10^{14}$ Hz, Yb $\approx5.2\times10^{14}$ Hz, magic λ ≈ 813 nm and ≈ 759 nm, $N\sim10^3$ – $10^4$ ) follow Ludlow et al. (2015).

Builds on. Dipole Force and the Optical Dipole Potential · The Recoil Limit and Sub-Recoil Cooling · Optical Molasses · Magneto-Optical Trap. Leads to. Relativistic Geodesy with Clocks · Quantum-Logic and Multi-Ensemble Clocks · Redefinition of the SI Second.

Ask the tutor: Ask me to walk through how the magic wavelength is measured in the lab, or to work a full accuracy budget from lattice depth to blackbody temperature for a strontium clock.

Does the magic wavelength make the trap completely perturbation-free?

No — it cancels only the leading term. The magic condition $\alpha_g=\alpha_e$ nulls the first-order, scalar differential light shift, but three effects survive. The hyperpolarizability shift scales as intensity squared, so it does not vanish at the linear null and must be measured by varying the lattice depth. The vector and tensor shifts depend on the lattice polarization and the atomic magnetic sublevel, so any ellipticity or field tilt reintroduces them. And the magic wavelength must be held to sub-picometre precision, because the differential polarizability still has a nonzero slope through the crossing. Reaching $10^{-18}$ means characterising all three, not assuming the trap is transparent.

Where does the photon recoil actually go on the carrier?

Into the lattice, and ultimately the whole apparatus. A free atom that absorbs a photon of momentum $\hbar k$ recoils, shifting its line by the recoil frequency $E_r/h$ — a few kilohertz, a fractional systematic near $10^{-11}$ that no clock could tolerate. A tightly bound atom cannot recoil freely: its motion is quantised, and on the carrier its vibrational quantum number is unchanged, so the kick has nowhere to go but the rigid lattice, exactly as a gamma recoil is absorbed by a crystal in the Mössbauer effect. Because the atom does not move, there is also no first-order Doppler shift.

Why not just use one atom, like an ion clock, and avoid collisions?

You can, and single-ion clocks are superb — they were the first optical clocks to reach $10^{-18}$ inaccuracy, precisely because a lone ion in an RF trap suffers almost no collisional or density shift. The price is statistics. Quantum projection noise falls as $1/\sqrt N$ , so a single particle averages roughly $\sqrt{10^4}=100$ times more slowly than a lattice of $10^4$ atoms at the same $Q$ . The lattice clock exists to recover that factor of 100. It pays with new problems — density-dependent collisions, lattice light shifts — which is why one-atom-per-site loading, fermionic Pauli suppression, and 3D lattices matter.

What sets the ultimate accuracy — is it the light shift or something else?

Once the magic wavelength is well maintained, the light shift is usually not the leader. In room-temperature clocks the blackbody-radiation shift dominates: a few parts in $10^{15}$ for Sr and, scaling as $T^4$ , demanding millikelvin knowledge of the radiative environment to correct to $10^{-18}$ . Close behind come the residual lattice shifts (hyperpolarizability, vector, tensor), density-dependent collisions, the DC-Stark shift from stray charges on nearby surfaces, and second-order Zeeman shifts from the bias field. Each is measured by modulating the responsible parameter — trap depth, atom number, field, temperature — and extrapolating to zero. The "accuracy budget" is literally a table of these effects with their evaluated uncertainties, summed in quadrature.

Why do lattice clocks use strontium and ytterbium in particular?

Both are alkaline-earth-like, with two valence electrons, and both have an extraordinarily narrow ¹S₀ → ³P₀ clock transition that is doubly forbidden and so intrinsically hair-thin — sub-millihertz natural linewidth in the fermionic isotopes ⁸⁷Sr and ¹⁷¹Yb, where hyperfine mixing gives just enough transition strength to drive it. Crucially, each has a convenient magic wavelength where high-power lasers exist: ≈ 813 nm for Sr, ≈ 759 nm for Yb. They also have simple level structures for laser cooling and calculable blackbody sensitivities. Other species (mercury, cadmium, magnesium) are studied too, often for a smaller BBR shift, but Sr and Yb hit the best overall balance of narrow line, coolability, and controllable systematics.

How does a lattice clock become a tool for geodesy?

Through the gravitational redshift: a clock deeper in a gravitational potential ticks slower, by a fraction $g\,\Delta h/c^2$ — about $1.1\times10^{-16}$ per metre or $1.1\times10^{-18}$ per centimetre near Earth's surface. A clock accurate at $10^{-18}$ can therefore sense a centimetre of elevation — or, read the other way, a difference in the geopotential — purely by comparing tick rates. Networks of clocks linked by stabilised fibre map geopotential differences between distant sites, complementing satellite gravimetry and spirit levelling. This "chronometric levelling" has already been demonstrated between labs and up mountains, turning the clock into a sensor of the shape of Earth's gravity field.